ServiceNow®
Predictive Intelligence is a platform function that provides a layer of
artificial intelligence that empowers features and capabilities across ServiceNow® applications to provide better work
experiences.
Note: Predictive Intelligence is also available for on-premise customers. If you're
interested in deploying this product on-premise, please contact your account manager. For
on-premise installation and configuration instructions, see Knowledge Base article
KB0782052.
Predictive Intelligence provides four frameworks that you can use to create
machine-learning solutions in your instance. Each framework delivers a different solution type
for training the system to predict, recommend, and organize data outcomes. A trained solution can
be invoked by any application through a prediction API to make a prediction. The frameworks
support these languages: English, French, German, Japanese, Dutch, Spanish, Italian, and
Brazilian Portuguese.
Note: Artificial intelligence communities use the term train
for machines in the same way we use it for humans and animals. For example, you can train horses
to jump higher, and you can train systems to learn new ways to process data and
solutions.
Predictive Intelligence classification framework
The Predictive Intelligence classification framework enables you to use
machine-learning algorithms to set field values during record creation, such as setting the
incident category based on the short description. You can train predictive models so they act as
an agent to automatically categorize and route work based on your past record-handling
experience.
Enable
Predictive Intelligence to handle later volumes of incoming requests at lower
costs. Automate the categorization and assignment of requests to reduce:
- Task resolution times.
- The number of interactions required to resolve tasks.
- The error rates of categorizing and assigning work.
For information on how to use the classification framework, see Create and train a classification solution.
Predictive Intelligence similarity framework
The Predictive Intelligence similarity framework identifies existing records that have
similar values to a new record. You use the framework to build a word corpus. The word corpus
functions as the vocabulary the system uses to compare your trained records based on their
textual similarity. For example, you can train a subset of your incident records to recommend a
resolution based on the information of a similar incident record. By reusing similar closed
incidents that have a proven resolution, you can help agents and fulfillers quickly provide the
best resolution for an incoming incident.
The similarity framework doesn't require an exact match of words for its text comparisons, as
its algorithms identify similar words and synonyms based on similar contexts. For example, the
phrases printer not working and printer broken are both captured in
your word corpus. The framework also collects, learns, and applies your industry-specific
context. For example, the phrase unable to join network has a different context in
a networking company than it does in a healthcare insurance company.
To keep your word corpus current, retrain and refresh your solution periodically. To see how
the Similarity Template helps you to configure your solution using similarity fields and
filters, see Create and train a similarity solution.
Predictive Intelligence clustering framework
Group similar records into clusters so you can address them collectively or identify patterns.
For example, you can group similar incidents that have occurred recently to identify a major
incident. To see how the Clustering Template helps you to configure your solution to identify
and train data for your cluster, see Create and train a clustering solution.
Predictive Intelligence regression framework
Regression is a machine-learning framework that you can train with historic data to predict
numeric outputs, such as a temperature or a stock price. For example, you can use regression to
estimate the time it takes to resolve an incident or a case. See Create and train a regression solution.
Training your machine-learning solutions
Predictive Intelligence enables you to train predictive models and machine-learning
solutions that you can apply to your business processes, such as:
You can also extend Predictive Intelligence to other processes by creating your own
predictive models and training them on your past record data.
Predictive model components
A predictive model consists of these components, some of which you must provide.
- Solution definition
- A data record you create and configure that specifies these values for training a
predictive model.
- The records used to train the model. For example, only train on incidents that are
resolved or closed within the last six months.
- The input fields the model uses to make predictions. For example, use the incident short
description to make a prediction.
- The output field whose value the model predicts. For example, set the incident category
based on the short description.
- The frequency to retrain the model. For example, retrain the model every 30 days.
- Solution
- The solution is the result of a solution definition that you've trained in a ServiceNow datacenter. Predictive Intelligence uses the solution
to predict a target field value given one or more input field values. All solutions specify
these values.
- The solution precision is the aggregate percentage of correct predictions.
For example, a precision of 50 means that out of 100 predictions, half of them should have
the correct value.
- The solution coverage is the aggregate percentage of records that receive a
prediction. For example, a coverage of 50 means half of all eligible records actually
receive a prediction.
- The solution classes are the output field values for which the model can
make predictions. Each class is an output field value with a list of possible precision,
coverage, and distribution metrics to choose from. For example, the Incident Categorization
solution has a class for each category such as software, inquiry, and database.
- The class distribution is the percentage of records from the entire table
that have this particular output field value. For example, a distribution of 50 for the
inquiry class means that half of incidents have the inquiry category.
- Business rule
- A rule that calls the solution data set to generate a prediction when a new record is
created.
Selecting data records for training your solution
A solution is only as good as the record data you use to train it. In general, a good training
dataset has these characteristics.
- The solution definition input fields are available to users when creating records. To make
predictions at record creation, the solution must have the input field values at record
creation.
- The solution definition output field is a choice field. To make more accurate predictions,
limit the output field to a finite set of possible values.
- The training records only contain correct values for the output field. To make more
accurate predictions, filter out any records that have unreliable output field values. For
example, if recently closed incidents are subject to review and change for a month, filter out
any recently closed incidents.
- The training records contain multiple examples of each output field value that you want the
solution to predict. To provide more record coverage, include multiple examples of each output
field value.
- The training records include common variations of the input fields. To provide more record
coverage, include multiple examples of input field values.
Exporting your solution for training
To train a solution, you export its solution definition and associated records to a
centralized training server within the nearest ServiceNow datacenter. When
the training completes, the training server exports the solution back to your instance and
deletes all of your training data from the server. As every datacenter has its own dedicated
training server and the data doesn't leave the datacenter, this service is also available to
customers who have data sovereignty requirements.
Predictions occur on a centralized prediction server within the same datacenter as the
instance. The trained model artifacts are sent from the instance server to the prediction server
when the prediction is invoked for the first time. After that, the trained model artifacts are
cached on the prediction server for subsequent predictions.
Note: All communication between the
instance and the training service occurs within the same datacenter firewall. Even so, all
communications occur over HTTPS.
Prediction business rules
By default, the system uses these business rules for
Predictive Intelligence.
| Name |
Description |
| Default Task Based Prediction |
A business rule that runs before inserting new task records to make a field value
prediction based on the solution definition output field and the solution dataset. Use this
business rule as a template to create your own prediction logic. This business rule calls
the Predictive Intelligence API. |
| Update Prediction Results |
A business rule that runs before closing task records to update the solution
statistics with the actual precision and coverage results. |
Monitoring your predictive model coverage and precision
You can track the coverage and precision of each predictive model using the Solution
Statistics dashboard, which provides reporting on these prediction areas by default.
| Report |
Description |
| Average Prediction Coverage (last 30 days) |
The value represents the percentage of predictions that yielded an outcome out of the
total number of predictions attempted. Click the coverage score to see a breakdown by
class. |
| Daily Prediction Coverage |
The value represents the percentage of records created on a given day in which the
solution was able to predict an outcome. |
| Average Prediction Precision (last 30 days) |
The value represents the percentage of predictions in which the predicted value was
the same as the final value of the field when the record closed. Click the precision score
to see a breakdown by class. |
| Daily Prediction Precision |
The value represents the percentage of records closed on a given day in which the
predicted field value was the same as the final value. |
For instructions on how to use the dashboard, see Review classification solution
statistics.
